TDT4205/ps5/src/generator.c

#include "vslc.h"

// This header defines a bunch of macros we can use to emit assembly to stdout
#include "emit.h"

// In the System V calling convention, the first 6 integer parameters are passed in registers
#define NUM_REGISTER_PARAMS 6
static const char* REGISTER_PARAMS[6] = {RDI, RSI, RDX, RCX, R8, R9};

// Takes in a symbol of type SYMBOL_FUNCTION, and returns how many parameters the function takes
#define FUNC_PARAM_COUNT(func) ((func)->node->children[1]->n_children)

static void generate_stringtable(void);
static void generate_global_variables(void);
static void generate_function(symbol_t* function);
static void generate_expression(node_t* expression);
static void generate_statement(node_t* node);
static void generate_main(symbol_t* first);

// Entry point for code generation
void generate_program(void)
{
  generate_stringtable();
  generate_global_variables();

  // This directive announces that the following assembly belongs to the .text section,
  // which is the section where all executable assembly lives
  DIRECTIVE(".text");

  // TODO (Task 3):
  // For each function in global_symbols, generate it using generate_function ()
  bool found = false;
  symbol_t *entry;
  for (int i = 0; i < global_symbols->n_children; i++) {
    symbol_t *sym = global_symbols->children[i];
    if (sym->type == SYMBOL_FUNCTION) {
      if (!found) {
        entry = sym;
        found = true;
      }
      generate_function(sym);
    }
  }

  // In VSL, the topmost function in a program is its entry point.
  // We want to be able to take parameters from the command line,
  // and have them be sent into the entry point function.
  //
  // Due to the fact that parameters are all passed as strings,
  // and passed as the (argc, argv)-pair, we need to make a wrapper for our entry function.
  // This wrapper handles string -> int64_t conversion, and is already implemented.
  // call generate_main ( <entry point function symbol> );
  generate_main(entry);
}

// Prints one .asciz entry for each string in the global string_list
static void generate_stringtable(void)
{
  // This section is where read-only string data is stored
  // It is called .rodata on Linux, and "__TEXT, __cstring" on macOS
  DIRECTIVE(".section %s", ASM_STRING_SECTION);

  // These strings are used by printf
  DIRECTIVE("intout: .asciz \"%s\"", "%ld");
  DIRECTIVE("strout: .asciz \"%s\"", "%s");
  // This string is used by the entry point-wrapper
  DIRECTIVE("errout: .asciz \"%s\"", "Wrong number of arguments");

  // TODO (Task 1): Print all strings in the program here, with labels you can refer to later
  // You have access to the global variables string_list and string_list_len from symbols.c
  for (int i = 0; i < string_list_len; i++) {
    DIRECTIVE("string%d:\t.asciz \"%s\"", i, string_list[i]);
  }
}

// Prints .zero entries in the .bss section to allocate room for global variables and arrays
static void generate_global_variables(void)
{
  // This section is where zero-initialized global variables lives
  // It is called .bss on linux, and "__DATA, __bss" on macOS
  DIRECTIVE(".section %s", ASM_BSS_SECTION);
  DIRECTIVE(".align 8");

  // TODO (Task 2): Fill this section with all global variables and global arrays
  // Give each a label you can find later, and the appropriate size.
  // Regular variables are 8 bytes, while arrays are 8 bytes per element.
  // Remember to mangle the name in some way, to avoid collisions with labels
  // (for example, put a '.' in front of the symbol name)

  // As an example, to set aside 16 bytes and label it .myBytes, write:
  // DIRECTIVE(".myBytes: .zero 16")
  for (int i = 0; i < global_symbols->n_children; i++) {
    symbol_t *sym = global_symbols->children[i];
    switch (sym->type) {
    case SYMBOL_GLOBAL_VAR:
      DIRECTIVE(".%s:\t.zero 8", sym.name);
      break;
    case SYMBOL_GLOBAL_ARRAY:
      node_t num = sym->node->children[1];
      assert(num->type == NUMBER_LITERAL);
      int len = num->data.number_literal * 8;
      DIRECTIVE(".%s:\t.zero %d", sym.name, len);
      break;
    case SYMBOL_FUNCTION:
      break;
    default:
      fprintf(stderr, "unexpected global symbol");
      exit(EXIT_FAILURE);
    }
  }
}

// Global variable used to make the functon currently being generated accessible from anywhere
static symbol_t* current_function;

// Prints the entry point. preamble, statements and epilouge of the given function
static void generate_function(symbol_t* function)
{
  // TODO (Task 3)

  // TODO (Task 3.1): Do the prologue, including call frame building and parameter pushing
  // Tip: use the definitions REGISTER_PARAMS and NUM_REGISTER_PARAMS at the top of this file
  LABEL(".%s", function->name);

  PUSHQ(RBP);
  MOVQ(RSP, RBP);

  symbol_table_t *symtable = function->function_symtable;
  for (int i = 0; i < symtable->n_symbols; i++) {
    symbol_t *sym = symtable->children[i];
    switch (sym->type) {
    case SYMBOL_PARAMETER:
      if (i < NUM_REGISTER_PARAMS) {
        PUSHQ(REGISTER_PARAMS[i]);
      } else {
        /* these params are already on the stack */;
      }
      break;
    case SYMBOL_LOCAL_VAR:
      PUSHQ("$0");
      break;
    default:
      fprintf(stderr, "unexpected symbol in symtable");
      exit(EXIT_FAILURE);
    }
  }

  // TODO (Task 4): the function body can be sent to generate_statement()
  generate_statement(function->node);

  // TODO (Task 3.2): Emit the epilogue, including a label and a default return value (0)
  LABEL(".%s.epilogue", function->name);
  MOVQ(RBP, RSP);
  POPQ(RBP);
  RET;
}

// Generates code for a function call, which can either be a statement or an expression
static void generate_function_call(node_t* call)
{
  // TODO (Task 4.3)
}

// Generates code to evaluate the expression, and place the result in %rax
static void generate_expression(node_t* expression)
{
  // TODO (Task 4.1): Generate code for evaluating the given expression.
  // (The candidates are NUMBER_LITERAL, IDENTIFIER, ARRAY_INDEXING, OPERATOR and FUNCTION_CALL)
  switch (expression->type) {
  case NUMBER_LITERAL:

    break;
  case IDENTIFIER:
    break;
  case ARRAY_INDEXING:
    break;
  case OPERATOR:
    break;
  case FUNCTION_CALL:
    break;
  }
}

static void generate_assignment_statement(node_t* statement)
{
  // TODO (Task 4.2):
  // You can assign to both local variables, global variables and function parameters.
  // Use the IDENTIFIER's symbol to find out what kind of symbol you are assigning to.
  // The left hand side of an assignment statement may also be an ARRAY_INDEXING node.
  // In that case, you must also emit code for evaluating the index being stored to
}

static void generate_print_statement(node_t* statement)
{
  // TODO (Task 4.4):
  // Remember to call safe_printf instead of printf
}

static void generate_return_statement(node_t* statement)
{
  // TODO (Task 4.5): Evaluate the return value, store it in %rax and jump to the function epilogue
}

// Recursively generate the given statement node, and all sub-statements.
static void generate_statement(node_t* node)
{
  if (node == NULL)
    return;

  // TODO (Task 4): Generate instructions for statements.
  // The statements you must handle are BLOCK, ASSIGNMENT_STATEMENT,
  // PRINT_STATEMENT, RETURN_STATEMENT and FUNCTION_CALL.
  // Statements of type LOCAL_VARIABLE should be ignored.
  for (int i = 0; i < node->n_children; i++) {
    node_t *child = node->children[i];
    switch (child->type) {
    case BLOCK:
      node_t list = child->children[0];
      for (int j = 0; j < list->n_children; j++)
        generate_statement(list->children[j]);
      break;
    case ASSIGNMENT_STATEMENT:
      generate_assignment_statement(child);
      break;
    case PRINT_STATEMENT:
      generate_print_statement(child);
      break;
    case RETURN_STATEMENT:
      generate_return_statement(child);
      break;
    case FUNCTION_CALL:
      generate_function_call(child);
      break;
    }
  }
}

static void generate_safe_printf(void)
{
  LABEL("safe_printf");

  PUSHQ(RBP);
  MOVQ(RSP, RBP);
  // This is a bitmask that abuses how negative numbers work, to clear the last 4 bits
  // A stack pointer that is not 16-byte aligned, will be moved down to a 16-byte boundary
  ANDQ("$-16", RSP);
  EMIT("call printf");
  // Cleanup the stack back to how it was
  MOVQ(RBP, RSP);
  POPQ(RBP);
  RET;
}

// Generates the scaffolding for parsing integers from the command line, and passing them to the
// entry point of the VSL program. The VSL entry function is specified using the parameter "first".
static void generate_main(symbol_t* first)
{
  // Make the globally available main function
  LABEL("main");

  // Save old base pointer, and set new base pointer
  PUSHQ(RBP);
  MOVQ(RSP, RBP);

  // Which registers argc and argv are passed in
  const char* argc = RDI;
  const char* argv = RSI;

  const size_t expected_args = FUNC_PARAM_COUNT(first);

  SUBQ("$1", argc); // argc counts the name of the binary, so subtract that
  EMIT("cmpq $%ld, %s", expected_args, argc);
  JNE("ABORT"); // If the provdied number of arguments is not equal, go to the abort label

  if (expected_args == 0)
    goto skip_args; // No need to parse argv

  // Now we emit a loop to parse all parameters, and push them to the stack,
  // in right-to-left order

  // First move the argv pointer to the vert rightmost parameter
  EMIT("addq $%ld, %s", expected_args * 8, argv);

  // We use rcx as a counter, starting at the number of arguments
  MOVQ(argc, RCX);
  LABEL("PARSE_ARGV"); // A loop to parse all parameters
  PUSHQ(argv);         // push registers to caller save them
  PUSHQ(RCX);

  // Now call strtol to parse the argument
  EMIT("movq (%s), %s", argv, RDI); // 1st argument, the char *
  MOVQ("$0", RSI);                  // 2nd argument, a null pointer
  MOVQ("$10", RDX);                 // 3rd argument, we want base 10
  EMIT("call strtol");

  // Restore caller saved registers
  POPQ(RCX);
  POPQ(argv);
  PUSHQ(RAX); // Store the parsed argument on the stack

  SUBQ("$8", argv);        // Point to the previous char*
  EMIT("loop PARSE_ARGV"); // Loop uses RCX as a counter automatically

  // Now, pop up to 6 arguments into registers instead of stack
  for (size_t i = 0; i < expected_args && i < NUM_REGISTER_PARAMS; i++)
    POPQ(REGISTER_PARAMS[i]);

skip_args:

  EMIT("call .%s", first->name);
  MOVQ(RAX, RDI);    // Move the return value of the function into RDI
  EMIT("call exit"); // Exit with the return value as exit code

  LABEL("ABORT"); // In case of incorrect number of arguments
  EMIT("leaq errout(%s), %s", RIP, RDI);
  EMIT("call puts"); // print the errout string
  MOVQ("$1", RDI);
  EMIT("call exit"); // Exit with return code 1

  generate_safe_printf();

  // Declares global symbols we use or emit, such as main and printf
  DIRECTIVE("%s", ASM_DECLARE_SYMBOLS);
}